Estimating prevalence: a confidence game.

Prevalence is one of the few estimates that rarely are reported with an appropriate measure of error in the parasitological literature. A minimum sample size recommendation of 15 samples, based on the relationship between sample size and standard error, likely has led to a false degree of confidence because of the nonlinear relationship between standard error and "true" 95% confidence intervals (as determined by Monte Carlo simulation or integration of the Bayesian posterior). Given that 95% confidence intervals for proportions are influenced by both sample size and the actual estimate of the proportion, there is no "gold standard" sample size beyond which estimates of binomial proportions can be considered "reliable." This necessitates the reporting of confidence interval estimates that have been shown to be conservative, such as the Clopper-Pearson estimate, or robust, such as the Wilson score approximation, or the computationally intensive integration of the Bayesian posterior.